Claude Opus 4.8 Launch: Better Judgment, Longer Autonomy, and Dynamic Workflows in Claude Code

Claude Opus 4.8 builds on Opus 4.7 with stronger judgment, more accurate self‑progress descriptions, and longer independent task execution. Price unchanged.

Fast mode

Fast mode uses the same model, delivers roughly 2.5× faster responses, and reduces the price to one‑third. Enable with /fast or request API access via http://claude.com/fast-mode.

Claude Code behavior

In Claude Code, Opus 4.8 can decide the next action without frequent user confirmation, stay on target during long conversations, and advance code‑base work toward a deliverable state. Users can hand over a feature, a bug‑fix round, or ongoing code cleanup to Claude and focus on other tasks.

Dynamic workflows (research preview)

Dynamic workflows let Claude first devise a plan, then run hundreds of parallel sub‑agents, validating results before reporting. Typical scenario: a large migration affecting hundreds of files.

Anthropic states that dynamic workflows enable end‑to‑end handling of the most challenging engineering tasks, compressing work that previously required quarterly planning into days. Claude writes orchestration scripts, schedules dozens to hundreds of sub‑agents in a single session, and performs self‑checks before delivering outcomes.

Targeted at problems too large for a single agent, especially complex legacy codebases such as service‑wide bug investigations, migrations across thousands of files, or pre‑deployment stress testing from multiple angles.

Available as a research preview for Claude Code CLI, Desktop, VS Code extension (Max, Team, Enterprise) and via Claude API, Amazon Bedrock, Vertex AI, Microsoft Foundry. Token consumption may be significantly higher than regular Claude Code sessions; start with well‑scoped tasks.

For the best experience, enable auto mode. Two ways to start a workflow:

Directly ask Claude to create a workflow, e.g., “Create a workflow”.

Enable the new setting ultracode in the effort menu, set effort level to xhigh ; Claude will automatically decide whether a workflow is needed.

Dynamic workflows in practice

Code‑base bug hunting, profiler‑guided optimization audits, and security audits: Claude searches the entire service or repository in parallel, validates each finding, and produces a focused report. The same flow can be used for code hardening, e.g., checking auth validation, input sanitization, and unsafe patterns across the codebase.

Large migrations and modernization: Claude can handle end‑to‑end framework replacements, API deprecations, and language porting across thousands of files.

Critical work requiring double‑check: When error costs are high, a workflow lets Claude attempt the same problem multiple times independently and deploy adversarial agents to proactively hunt for vulnerabilities before delivering results.